Results of the mechanistic substudy

This chapter describes the mechanistic substudy of the GaPP2 clinical trial.

Introduction

Increasingly, chronic pain conditions are being acknowledged to be associated with a number of central nervous system changes that may be responsible for generating or maintaining the pain.8 Changes within key regions, including somatosensory, emotion regulation and components of the descending pain modulatory system (DPMS) [including the anterior cingulate cortex (ACC), amygdala, periaqueductal grey matter (PAG) and rostral ventromedial medulla], have been demonstrated both in conditions with a known peripheral pathology (e.g. osteoarthritis,52–54 shoulder impingement syndrome,55 endometriosis-associated pain)56,57 and in those for which no pathology can be found (e.g. fibromyalgia,58,59 chronic back pain,60 irritable bowel syndrome61 and somatoform pain disorder62). CPP appears to be no different, with reviews63,64 identifying similar changes in women with and without identified pathology. However, more is known about irritable bowel syndrome and internal cystitis/bladder pain syndrome than about laparoscopy-negative CPP when considering pain syndromes without identified pathology. Studies of CPP relatively consistently demonstrate widespread hyperalgesia (an increased response to a stimulus that was previously painful) and often allodynia (pain from a previously non-painful stimulus). The neuroimaging studies that have explored the correlates of these sensations implicate a variety of regions (as with other chronic pain conditions), including those involved with emotion regulation and the PAG, as potentially involving dysfunctional pain regulatory mechanisms. However, to date, there are no neuroimaging studies focusing on women with CPP of unknown cause. In addition, women with CPP from a variety of causes have been shown to have hypothalamic–pituitary–adrenal axis dysfunction, autonomic system changes and, frequently, psychological distress.63

Gabapentin is a neuromodulator for which there is evidence of efficacy in neuropathic pain conditions,65 such as postherpetic neuralgia66 and diabetic neuropathy.67 In addition, it is also used for conditions in which there is considered to be a central component to the pain (e.g. fibromyalgia). Several animal studies have shown reductions in central sensitisation following intrathecal administration of gabapentin.68–71 In humans, an important element of gabapentin’s efficacy appears to be supraspinal, as intrathecal administrations are insufficient to induce analgesia.72 In anaesthetised rats, gabapentin induces a dose-dependent increase in thalamic and PAG activation, as well as decreased activation of the limbic system (amygdala and entorhinal cortex).35 Neuroimaging work in healthy humans has found that a single 1800-mg dose of gabapentin during experimentally-induced central sensitisation reduced both PAG activation and pain-induced deactivations,33 as well as insula and mesencephalic reticular formation activation.73 This suggests that gabapentin could constitute an effective ‘antihyperalgesic’, which supports its use in chronic pain syndromes. However, it remains to be established whether or not gabapentin is efficacious in CPP17 or what factors may contribute to the likelihood of its success in treating any particular person.

Randomised controlled trials of prospective analgesics always face the hurdle of strong placebo effects. It has been proposed that we need a more mechanistic understanding of how analgesics engage pain networks, and to relate this to symptom change within the individual, to establish true efficacy.74 Neuroimaging techniques offer a non-invasive and objective means to assess how pain networks are altered over time in the presence of a particular treatment. Within this RCT of gabapentin for CPP, we embedded a pre- and post-treatment fMRI study that examined brain responses to punctate stimuli. The aims were to determine (1) whether or not gabapentin, in comparison with placebo, altered brain responses to pain; (2) whether or not this covaried with positive clinical responses over time; and (3) whether or not pre-treatment brain responses could predict positive clinical response. In addition to measures of subjective pain, we were concerned with our participants’ mental well-being and ability to function in day-to-day life; a series of participant-reported validated questionnaires encompassing these domains were, therefore, collected at baseline and after 16 weeks of treatment.

Objectives

• Determine the presence of central nervous system changes in women with CPP and no obvious underlying pathology.
• Determine the effect of gabapentin on central pain processing in women with CPP and no underlying pathology.
• Determine whether or not there are baseline fMRI measures that correlate with response to treatment.
• Determine whether or not there are clinical measures that correlate with response to treatment.

Methods

Participants were drawn from those who were recruited to the GaPP2 trial from the Edinburgh region. In total, 45 women underwent a pre-treatment baseline MRI scan, and 25 women returned for a second scan after 12 weeks of maximum-tolerated dose of the intervention (gabapentin or placebo). Figure 4 demonstrates group allocation and reasons for exclusion. In addition to the inclusion and exclusion criteria of the main trial, additional exclusion criteria applied to those undergoing MRI: the presence of implants not compatible with magnetic resonance, such as pacemakers, and claustrophobia.

FIGURE 4 Participant group allocation and reasons for exclusion from the study analysis. GnRH, gonadotropin-releasing hormone; HRT, hormone replacement therapy. (Figure not reproduced here; see the original article.)

Sample size

The sample size was based on data from our pilot study, GaPP1.1 In that study, 12 women had a fMRI scan after the 12 weeks of treatment, with 11 usable data sets. Of these, five were using placebo and six were using gabapentin. A priori region of interest (ROI) analysis demonstrated a > 1.2% difference in the blood oxygen level-dependent signal in the PAG and a > 1.4% difference in the left posterior insula between the groups. With a power of > 80% and p = 0.05, we, therefore, calculated that we would need to recruit 50 women (25 per treatment group) to detect a difference in signal from key ROIs. This sample size is greater than that used in a number of other pharmacological fMRI studies.

Magnetic resonance imaging scanning

Scanning took place at the Edinburgh Imaging Facility, Queen’s Medical Research Institute,75 using a 3T Siemens Magnetom Verio scanner (Siemens Healthcare GmbH, Erlangen, Germany). Scanning constituted a high-resolution structural scan; functional imaging during both a resting state scan and the application of punctate stimuli; field map acquisitions; and pseudo-continuous arterial spin labelling. The primary outcome of interest for the mechanistic study was functional responses to punctate stimuli, and this is the focus of this chapter. Future analyses will incorporate data from the remaining sequences acquired.

The blood oxygen level-dependent functional images were acquired using an echoplanar T2* GRAPPA (generalised autocalibrating partial parallel acquisition) gradient echo pulse sequence, with a repetition time (TR) of 2500 milliseconds, an echo time (TE) of 30 milliseconds, a flip angle of 90°, a field of view (FOV) of 192 mm, 45 interleaved contiguous slices and a resolution of 3 × 3 × 3 mm. A T1-weighted structural image was acquired using a MPRAGE (magnetisation-prepared rapid gradient echo) sequence, with a TR of 2300 milliseconds, a TE of 2.98 milliseconds, a FOV of 256 mm, a resolution of 1 × 1 × 1 mm and a flip angle of 9°. The punctate task was 10 minutes and 50 seconds, that is 260 volumes, the first four of which were discarded to reduce T1 saturation effects. The resting state scan lasted 8 minutes and 20 seconds, with 200 volumes. The resting state scan was performed prior to the punctate scan to avoid flaring participants’ pain and, thus, not truly measuring a resting state. Gradient field maps were acquired with the same dimensions as the functional data, and TE1 (echo time 1) 4.92 milliseconds, TE2 (echo time 2) 7.38 milliseconds. The pre-treatment scan occurred at visit 1 of the trial schedule, and the post-treatment scan between week 13 and week 16 (visit 5 of the trial schedule). Both scans were identical and for each participant took place during the same phase of their menstrual cycle.

The punctate scan involved the application of 39 punctate stimuli using a 300 g Touch Test von Frey filament (Ugo Basile, Gemonio, Italy) (6.65 mm) in an event-related design. These were applied by five researchers who had undergone appropriate training and were directed by timed auditory cues. The mean interstimulus interval was 16 seconds, with random jitters of 2.5 seconds. Stimuli were applied to the lower abdomen, 10 cm above the superior edge of the pubic bone. In contrast to the original grant application, pain processing at a site distant to the clinical pain was not carried out owing to the limitation of the equipment used for thermal stimulation of the left hand. Use of the thermal stimulation equipment in the magnetic resonance imaging (MRI) scanner created significant noise, which lead to distortion of the fMRI signal that was not correctable with analysis tools. We attempted to reduce this noise by altering filters between the control and the scanner rooms, but were unable to make a meaningful improvement. Thus, the thermal stimulation component of the paradigm was removed. As this was at a control site rather than the site of referral of pain, it was not the primary fMRI sequence of interest and, thus, the key data we had aimed to collect are still available (evoked pain from a referral site and resting state data).

Outcome measures

• fMRI measures of evoked pain.
• fMRI measures of resting brain activity and connectivity.
• Post-treatment minus pre-treatment difference in fMRI measures.
• Correlation of participant-reported outcomes and fMRI measures.

Statistical analysis plan

The SAP for the mechanistic substudy was agreed between the research team, including the a priori ROI and whole-brain analyses previously described. Further exploratory analyses of this rich data set will continue, to improve our understanding of the main mechanistic findings.

Covariates of interest

An a priori selection of participant-reported secondary outcomes was drawn from the clinical trial and incorporated into a series of exploratory analyses of fMRI data. NRS pain scores were taken from the pre-treatment and post-treatment assessments for average and worst pain. Physical and mental well-being measures were used from the Short Form Health Survey (SF-12) (with the mental component score and physical component score). The degree to which pain interfered with daily living was taken from the BPI. Fatigue and psychological distress were taken from the BFI and GHQ-12, respectively. Levels of rumination, pain magnification and helplessness were taken from the PCS. Neuropathic pain scores were taken from PainDETECT.

Punctate functional magnetic resonance imaging data analysis

The principal analysis of the punctate fMRI data took place within Statistical Parametric Mapping (SPM12) software [version 12 (7771); The Wellcome Centre for Human Neuroimaging, University College London, London, UK]. Data were unwarped using the field maps and realigned to a mean image, which was co-registered to each person’s T1-weighted structural image. This was, in turn, segmented and warped to the Montreal Neurological Institute (MNI) space, with warp parameters applied to the co-registered functional images. The data were finally smoothed with a 5 mm full width half maximum kernel. Quality assurance procedures examined the data for slice spikes and significant motion (defined as a single-volume motion exceeding half a voxel, i.e. 1.5 mm). Volumes that demonstrated problems were replaced with a dummy volume interpolated from those acquired just before and after it, and such volumes were represented within a nuisance regressor within the affected participants’ first-level design matrices. If data contained more than 10 significant movement events, the participant was excluded.

For each participant, the onset of each punctate stimulus was represented as a delta function and was convolved with the canonical haemodynamic response function, as well as its temporal and dispersion derivatives. Nuisance regressors representing motion parameters (and dummy volumes where applicable) were included, and the contrast for the canonical punctate regressor was taken forward into random-effects flexible-factorial analyses. This modelled the main effects of participant, time, treatment and the treatment group by time interaction. Incorporating the main effects of the treatment and participant would in part account for any nuisance variance between the groups at baseline. The main effects of treatment, time and the interaction were assessed, with clusters being reported as significant if they achieved a whole-brain family-wise error rate (FWE) corrected threshold of p < 0.05. To enhance sensitivity within the brainstem, a further analysis was restricted to within an anatomical mask of the midbrain, pons and medulla, as defined by Brodmann,76 and implemented within the WFU Pickatlas software (version 3.0.5; Wake Forest University School of Medicine, Winston-Salem, NC, USA).77,78

For any regions demonstrating significant treatment-by-time interactions, the impact that this had on our covariates of interest was assessed in a series of general linear models. The first eigenvariate was extracted from significant clusters and analysed in SPSS version 24.0 (IBM SPSS Statistics for Macintosh, IBM Corporation, Armonk, NY, USA). Each participant-reported measure was the dependent variable, treatment group was a random factor and the changes in brain activation over time were entered as covariates, with covariate-by-treatment interactions being modelled.

Finally, to determine whether or not punctate activation at baseline was able to predict subsequent clinical response, the post-treatment minus pre-treatment difference for each participant-reported measure was analysed within a general linear model using SPSS. This modelled treatment as a random factor, and the pre-treatment punctate-induced responses from those regions demonstrating a treatment-by-time interaction were included as covariates.

Results

Behavioural measures

Punctate stimuli were perceived as painful by some women. Pain intensity ratings after the punctate paradigm ranged from 0 to 8 (pre-treatment scan: 3.56 ± 2.36; end-of-treatment scan: 3.12 ± 2.15). There was no significant change in pain intensity of these stimuli between the two scans for either group, and there was not a time × treatment interaction [placebo: scan 2 > scan 1, t(11) = –0.764, p = 0.461; gabapentin: scan 2 > scan 1, t(11) = –0.672, p = 0.515; time × treatment interaction F(1,22) = 0.006, p = 0.938].

The effects of punctate stimuli

To demonstrate validity of the task, the main effects of punctate stimuli across both treatment groups and visits are displayed in Table 19 and Figure 5. This demonstrated activation and deactivation of regions typically associated with central pain processing, including bilateral insula, thalamus, ACC and somatosensory cortices.

FIGURE 5 Axial slices displaying whole-brain significant punctate-induced activation and deactivation. Displayed clusters have p < 0.05 FWE-corrected for the whole-brain volume. (Figure not reproduced here; see the original article.)

Punctate validation analyses

The main effects of time (across both treatment groups) and the effects of treatment at time 1 (post allocation but pre intervention) were analysed. A lack of significant findings for these analyses would support the validity of the data as a whole. There were no main effects of time that achieved statistical significance (p > 0.551 FWE-corrected). Comparing the two treatment groups prior to treatment showed one region of significant difference: the left postcentral gyrus [Brodmann area = 40; MNI –36, –26, 74; peak z = 4.98; kE (cluster extent) = 134; cluster p < 0.001 FWE-corrected], which was not part of the primary somatosensory cortex.

Punctate: treatment-by-time interaction

The critical contrast of interest was to establish whether or not gabapentin altered brain activity over time compared with changes seen in the placebo group. The treatment-by-time interaction showed significant effects in the ACC (pregenual and anterior midcingulate subdivisions) and cuneus (Table 20 and Figure 6). Data were extracted from these regions, and post hoc t-tests confirmed that pain-evoked activity within the ACC significantly decreased following treatment with gabapentin [t(12) = –5.763; p < 0.001] and increased in the placebo group [t(11) = 3.784; p = 0.003]. In the cuneus, activation was significantly decreased in the gabapentin group only [t(12) = 5.126, p < 0.001, vs. p = 0.204 for placebo]. These results remained highly significant after covarying for baseline average pain scores within a repeated-measures analysis of covariance (p < 0.001). No areas demonstrated a significant interaction in the brainstem mask. These findings remained significant when the pain intensity ratings obtained after the punctate paradigm were included as a covariate.

FIGURE 6 Treatment-by-time interaction punctate activation, and associated clinical correlations. (a) Sagittal slices depict the ACC and cuneus, from top to bottom. Both clusters are p < 0.05 FWE-corrected for the whole-brain volume; (b) extracted activation displaying the nature of the treatment-by-time interactions, ** denotes post hoc t-test significance of < 0.005. Error bars are 95% CI; (c) scatterplots demonstrating the relationships between changes in BPI pain interference scores and ACC activation. For BPI pain interference score, negative changes over time indicate an improvement in symptoms. (Figure not reproduced here; see the original article.)

Covariance with clinical measures of interest

The treatment-by-time interaction in ACC was related to a significant improvement in BPI pain interferences scores [F(1,17) = 12.905; p = 0.002; see Figure 6c). For women in the gabapentin group, those who showed larger reductions in ACC activation post treatment also had the greatest improvements in their pain interference scores (Pearson’s r = 0.562; p = 0.046). The placebo group demonstrated a markedly different pattern: the more their ACC response increased over time, the greater their improvement on the BPI pain interference score (Pearson’s r = –0.605; p = 0.037). Cuneus changes were not found to differentially impact on clinical improvement.

Punctate: predicting a positive response from baseline data

The ACC activation at baseline significantly predicted improvements in the physical component of the SF-12 across both groups [physical component score, F(1,17) = 9.341; p = 0.007], but also demonstrated a significant group interaction [F(1,17) = 5.452; p = 0.032]. Figure 7a suggests that, although those with elevated pre-treatment ACC activation appear to improve in general, the effect is especially pronounced in the gabapentin group.

FIGURE 7 Predicting response to gabapentin using pre-treatment punctate activation. (a) SF-12 physical well-being score; and (b) PainDETECT score. (Figure not reproduced here; see the original article.)

The ACC responses at baseline also predicted improvements in neuropathic pain scores as measured by PainDETECT [F(1,17) = 7.142; p = 0.016]. The pattern was similar to that seen in the physical component score (see Figure 7b) in that elevated pre-treatment ACC predicted better outcomes, but this was much more marked in those taking gabapentin.

In all cases, the black line represents the fit across both groups. For those in the gabapentin group, the greatest improvements in (1) physical well-being (i.e. increases in SF-12 physical component scores) and (2) neuropathic pain (i.e. decreased PainDETECT scores) are seen in those with elevated pre-treatment ACC responses. However, bivariate correlations between the pre-treatment ACC activation and the clinical measures at baseline did not show a significant positive relationship with PainDETECT scores (r = 0.374; p = 0.066).

To examine this further in the clinical trial’s primary outcome measures, the study sample was split according to the median scores of average pain (post treatment minus pre treatment). Within the gabapentin group, those who showed the most improvement in average pain had elevated baseline ACC activation [t(11) = 2.309; p = 0.041] and a more pronounced reduction in ACC activation after treatment [t(11) = 2.664; p = 0.022]. This was not seen in the placebo group (p > 0.314). There were no such effects in the cuneus or for worst pain scores.

Discussion

In the mechanistic group of the GaPP2 trial, we have shown that a 12-week course of gabapentin appears to exert an effect on the ACC in women with CPP. Those taking gabapentin show a significant reduction in ACC responses to punctate stimuli after treatment, which in turn was associated with improvements in pain interference scores. ACC activity at baseline also predicted those most likely to respond to gabapentin, with participants who had the most ACC activation in response to punctate stimuli showing the largest improvements in physical well-being and neuropathic pain. Although these findings are in a small cohort only, the ability of neuroimaging to detect a signal of efficacy and disentangle treatment effects from placebo is one of the strengths of a design such as this.

The ACC is known to play a role in pain processing8 and is a critical component of the DPMS.79 In humans, pain responses during experimentally induced central sensitisation33 and multiple chronic pain conditions53,60,61,80 are associated with increased rostral ACC activation. The ACC regions demonstrating a treatment-by-time interaction encompassed both the most rostral aspects of the anterior midcingulate cortex, and the pregenual subdivision. In healthy volunteers, these regions mediate placebo analgesia81 in part via mu opioid receptors.82 Interestingly, infusion of the rostral ACC with gabapentin in rats with spinal nerve ligation reduced the aversive aspects of pain and facilitated pain relief-motivated behaviours. This effect appeared to be dependent on endogenous opioid signalling, and was not associated with reduced tactile allodynia itself.83 Others have proposed that pharmacological modulation of the ACC may allow for a reduction of the distress associated with chronic pain, without impairing the physiological function of acute pain.84 This would be consistent with the regions known roles in processing negative emotions,85 resolving conflict86 and goal selection/maintenance.87 It is proposed that the ACC integrates the aversiveness of pain to guide adaptive avoidant behaviours.85 It is, therefore, significant that we observe gabapentin reducing ACC responses to pain in a way that correlates with improvements in pain-related interference in daily living.

We also demonstrated a significant treatment-by-time interaction in the cuneus, although it is less clear what role it may be playing. Its perfusion has been shown to be altered in migraine sufferers, in a manner inversely related to induced pain scores.88 It also appears to relate to the affective appreciation of pain, with pain-induced responses correlating with affective ratings in healthy volunteers.89 Here, we found no clinical correlations.

Here, we present evidence that gabapentin acts to alter function within the DPMS, the degree to which correlates with clinical improvement. However, we did not observe significant treatment-by-time interactions in other DPMS brainstem structures, such as the PAG or rostral ventromedial medulla. It may be that this study was underpowered to detect such changes, or that assessing simple differences in activation did not capture the relationship between these structures and other DPMS regions. Alternatively, it may be that the regions in the brainstem are too small to be detected with whole-brain/brainstem analyses.

The evoked data are the most meaningful with regard to understanding the mechanism of drug activity. The resting state data will be explored to understand this mechanism in more detail (informed by the results of the evoked analysis, i.e. specific ROIs to investigate) and to further understand a global impact of gabapentin, and how it potentially generates adverse effects.

Previous imaging studies of gabapentin in chronic pain have used a single pre-scan dose,73,90 and, as far as we are aware, this study is the first to perform longitudinal neuroimaging pre treatment and post treatment in a cohort of women with established CPP. To date, most pharmacological functional imaging studies have involved male participants, in part because of concerns regarding controlling for hormonal variation and also to avoid exposure to potentially teratogenic effects of medication. Therefore, significant strengths of this study are that it involved a cohort of immediate relevance to CPP and that all women underwent their post-treatment scan at a similar point in their menstrual cycle to the baseline scan. This study has also demonstrated the utility of incorporating pre-treatment and post-treatment neuroimaging into a clinical trial of pain medication; we can present evidence that gabapentin acts on chronic pain by modulating the DPMS system via ACC (in contrast to other mechanisms implicated in CPP).63 We can also suggest that a subgroup of women, those demonstrating pronounced ACC responses to punctate stimulation, are more likely to benefit from taking gabapentin than others. The challenge remains to identify preclinical factors correlating with these enhanced ACC responses that would allow responders to be identified without the need for neuroimaging. The neuropathic PainDETECT score shows promise; however, no significant correlation with baseline ACC activation was identified.